Remote Sensing of Environment 193 (2017) 257–273

Contents lists available at ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r .com/ locate / rse

Evaluating soil moisture retrievals from ESA's SMOS and NASA's SMAPbrightness temperature datasets

A. Al-Yaari a,⁎, J.-P. Wigneron a, Y. Kerr b, N. Rodriguez-Fernandez b, P.E. O'Neill c, T.J. Jackson d,G.J.M. De Lannoy e, A. Al Bitar b, A. Mialon b, P. Richaume b, J.P. Walker f, A. Mahmoodi b, S. Yueh g

a INRA, UMR1391 ISPA, Villenave d'Ornon, Franceb CESBIO, Université de Toulouse, CNES/CNRS/IRD/UT3, UMR 5126, Toulouse, Francec NASA, Goddard Space Flight Center, Greenbelt, USAd USDA-ARS Hydrology and Remote Sensing Laboratory, Beltsville, USAe KU Leuven, Department of Earth and Environmental Sciences, Heverlee, Belgiumf Department of Civil Engineering, Monash University, Clayton, Melbourne, Victoria, Australiag Jet Propulsion Lab., California Inst. of Technol., Pasadena, CA, USA

⁎ Corresponding author.E-mail address: amen.al-yaari@inra.fr (A. Al-Yaari).

http://dx.doi.org/10.1016/j.rse.2017.03.0100034-4257/© 2017 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 12 May 2016Received in revised form 6 March 2017Accepted 12 March 2017Available online xxxx

Two satellites are currently monitoring surface soil moisture (SM) using L-band observations: SMOS (Soil Mois-ture and Ocean Salinity), a joint ESA (European Space Agency), CNES (Centre national d'études spatiales), andCDTI (the Spanish government agency with responsibility for space) satellite launched on November 2, 2009and SMAP (Soil Moisture Active Passive), a National Aeronautics and Space Administration (NASA) satellite suc-cessfully launched in January 2015. In this study, we used a multilinear regression approach to retrieve SM fromSMAP data to create a global dataset of SM, which is consistent with SM data retrieved from SMOS. This wasachieved by calibrating coefficients of the regression model using the CATDS (Centre Aval de Traitement desDonnées) SMOS Level 3 SM and the horizontally and vertically polarized brightness temperatures (TB) at 40° in-cidence angle, over the 2013–2014 period. Next, this model was applied to SMAP L3 TB data from Apr 2015 to Jul2016. The retrieved SM from SMAP (referred to here as SMAP_Reg) was compared to: (i) the operational SMAPL3 SM(SMAP_SCA), retrievedusing thebaseline Single Channel retrieval Algorithm(SCA); and (ii) the operation-al SMOSL3 SM, derived from the multiangular inversion of the L-MEB model (L-MEB algorithm) (SMOSL3). Thisinter-comparison was made against in situ soil moisture measurements from N400 sites spread over the globe,which are used here as a reference soil moisture dataset. The in situ observations were obtained from the Inter-national Soil Moisture Network (ISMN; https://ismn.geo.tuwien.ac.at/) in North of America (PBO_H2O, SCAN,SNOTEL, iRON, and USCRN), in Australia (Oznet), Africa (DAHRA), and in Europe (REMEDHUS, SMOSMANIA,FMI, and RSMN). The agreement was analyzed in terms of four classical statistical criteria: Root Mean SquaredError (RMSE), Bias, Unbiased RMSE (UnbRMSE), and correlation coefficient (R). Results of the comparison ofthese various products with in situ observations show that the performance of both SMAP products i.e.SMAP_SCA and SMAP_Reg is similar and marginally better to that of the SMOSL3 product particularly over thePBO_H2O, SCAN, and USCRN sites. However, SMOSL3 SM was closer to the in situ observations over theDAHRA and Oznet sites. We found that the correlation between all three datasets and in situ measurements isbest (R N 0.80) over the Oznet sites and worst (R = 0.58) over the SNOTEL sites for SMAP_SCA and over theDAHRA and SMOSMANIA sites (R= 0.51 and R=0.45 for SMAP_Reg and SMOSL3, respectively). The Bias valuesshowed that all products are generally dry, except over RSMN, DAHRA, and Oznet (and FMI for SMAP_SCA). Fi-nally, our analysis provided interesting insights that can be useful to improve the consistency between SMAPand SMOS datasets.

© 2017 Elsevier Inc. All rights reserved.

Keywords:SMOSSMAPSoil moistureStatistical regression

1. Introduction

Lately, the importance of soil moisture has become increasingly ap-parent, because soilmoisture is a key variable in better understanding ofthe land-atmosphere interactions (Chen et al., 2016; Hirschi et al.,

258 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

2014). The exchange of heat andwater between the land surface and at-mosphere is influenced by soil moisture (Berg et al., 2014; Hupet &Vanclooster, 2002; Seneviratne et al., 2010; Western et al., 2004),which was recognized as an Essential Climate Variable (ECV) in 2010(GCOS, 2010).

Global soil moisture information has become available via differentactive and passive microwave remote sensing techniques with goodtemporal and spatial resolutions (Bartalis et al., 2007; Kerr et al., 2001;Njoku et al., 2002; Njoku et al., 2003; Owe et al., 2001; Ulaby et al.,1996; Wigneron et al., 1995). However, the required temporal and spa-tial resolutions strongly depend on the applications (e.g., agriculturalapplications vs. climate studies). Recently, new global soil moisturedatasets, with a typical target accuracy of 0.04 m3/m3 (Jackson et al.,2016; Kerr et al., 2010; Kerr et al., 2012) over bare, low vegetationcover, and sparsely vegetated areas, have been produced based on mi-crowave satellite observations at L-band (1.4 GHz, 21 cm). L-band isconsidered optimal for soil moisture monitoring (Kerr et al., 2001;Njoku et al., 2003;Wang & Choudhury, 1981) due to its higher sensitiv-ity to soil moisture and penetration into vegetation and soil (Kerr, 2007;Njoku et al., 2003; Owe & Van de Griend, 1998; Wang & Choudhury,1981) than other higher frequencies (e.g., C-band, X-band, etc.). Thenew L-band based datasets include surface soil moisture from twospaceborne missions: ESA's (European Space Agency) Soil Moistureand Ocean Salinity (SMOS) (Kerr et al., 2012) and NASA's (NationalAeronautics and Space Administration) Soil Moisture Active Passive(SMAP) (Entekhabi et al., 2010). The SMOS and SMAP satellites werelaunched in 2009 and 2015, respectively, and have been providing mi-crowave brightness temperature (TB) observations since then. Soilmoisture information is retrieved from SMAP's and SMOS's TB observa-tions based on the principle that soil TB is mainly determined by soilmoisture via soil dielectric constant (Njoku et al., 2002; Schmugge etal., 1976; Ulaby et al., 1996). Nevertheless, the sensitivity of the SMOSand SMAPTB observations to soil moisture is reduced by perturbing fac-tors such as vegetation (attenuation of the emission from the soil andadditional upwelling emission toward the space-borne sensor), surfaceroughness (scattering effects increase the emitting surface area), topog-raphy, soil texture, soil bulk density, and soil temperature (Choudhuryet al., 1979; Grant et al., 2008; Holmes et al., 2006; Jackson &Schmugge, 1991; Kerr et al., 2012; Njoku & Li, 1999; Njoku et al.,2003; Wang et al., 1983; Wigneron et al., 2007; Wigneron et al., 2011;Wigneron et al., 2017).

There are several remotely sensed soil moisture products available(in addition to SMOS and SMAP); however, these cover different pe-riods and are not consistent in terms of spatial and temporal resolutions,period availability, grid, etc. Given the wide availability of soil moisturedatasets retrieved from different microwave observations, studies fo-cusing on the merging of these products are important to advance inthe field of producing long-term and consistent datasets of several cli-matic variables. A great effort has beenmade by the scientific communi-ty in the last decade to build a coherent and consistent long term soilmoisture datasets such as the ESA Climate Change Initiative (CCI) soilmoisture data record (e.g., Enenkel et al., 2015; Liu et al., 2012; http://www.esa-soilmoisture-cci.org/; Wagner et al., 2012), deemed neces-sary for global soil moisture monitoring, drought monitoring, climateforecasts, etc. The CCI product is estimated based on a posteriori merg-ing i.e. merging the retrieved soil moisture datasets based on the rela-tive errors of soil moisture products and a CDF (cumulativedistribution function)-matching used to rescale the different soil mois-ture products into a common climatology. An alternative approach isto use data fusion i.e. merging of microwave datasets prior to the re-trieval (e.g., through the use of a common retrieval algorithm as pro-posed later in this paper). This method allows better exploitation ofthe complimentary of information provided by the different sensorsnot included in the posteriori combination approach (Aires et al.,2012; Kolassa et al., 2013). A recent project was established by ESA toinvestigate the integration of SMOS soil moisture estimates within the

CCI soil moisture data record using three approaches that implementthe data fusion strategy:

(i) multi-linear regression (Al-Yaari et al., 2016);(ii) neural networks (Rodríguez-Fernández et al., 2016); and(iii) the Land Parameter Retrieval Model (LPRM; Van der Schalie et

al., 2016).

Al-Yaari et al. (2016), for instance, demonstrated the efficiency ofphysically-based multiple-linear regression equations (Wigneron etal., 2004), referred to here as Linear RegressionMethod (LRM) in the fol-lowing, to retrieve soil moisture from the Advanced Microwave Scan-ning Radiometer Earth Observing System (AMSR-E) TB observations.The LRM has several advantages: quickness, simplicity, and no strongdemand on auxiliary datasets (Al-Yaari et al., 2016) such as the normal-ized difference vegetation index (NDVI) product used by the SMAP Sin-gle Channel Algorithm (SMAP_SCA), to estimate vegetation effects. Thepurpose of that initial studywas to extend the SMOS soil moisture prod-uct into the past i.e., 2003–2009, using AMSR-E TB observations. Thecurrent study follows the same strategy to retrieve soil moisture fromSMAP TB observations (SMAP_Reg) with a purpose to improve the tem-poral sampling rate togetherwith the SMOS soilmoisture product at theglobal scale. Themain interest in the SMAP-Reg soil moisture product isthat it is fully consistent (coherent in temporal dynamics and absolutevalues) with the SMOS Level 3 soil moisture product, as the regressionequations are calibrated based on SMOS Level 3 data (soil moistureand TB). Furthermore, the idea here is to re-build a coherent and consis-tent soil moisture dataset rather than to develop a new algorithm or tosurpass the well-established radiative transfer models (e.g. the L-bandMicrowave Emission of the Biosphere (L-MEB) model, LRPM, etc.).

To this end, two specific objectives of this study are listed below:

(i) produce a soil moisture product (SMAP_Reg) from SMAP TB thatis consistentwith SMOS soil moisture retrievals using physically-based regression equations; and

(ii) compare SMAP_Reg with operational SMAP and SMOS soil mois-ture retrievals against ground-based soilmoisturemeasurements.

Since SMAP soil moisture products are relatively recent, their evalu-ation and their inter-comparison with other soil moisture datasets arerequired (Chan et al., 2016; Zeng et al., 2016). To advance our goal,therefore, the second objective of this study is two-fold: to evaluatethe SMAP_Reg product, and to carry out a first evaluation of the agree-ment between SMAP and SMOS Level 3 soil moisture products on aglobal scale and against ground-based measurements (sparse anddense networks). The aim is not to establish which product is more ac-curatewith respect to in situ but to understand the spatio-temporal pat-terns of SMAP relative to SMOS and how SMAP differs from SMOSglobally. The agreement and degree of dispersion between the SMAPand SMOS soil moisture products are analyzed here in terms of fourclassical statistical criteria: Root Mean Squared Error (RMSE), Bias, Un-biased RMSE (UnbRMSE), and correlation coefficient (R) during theoverlapping period (from Apr 2015 to Jul 2016).

The datasets, the local regression method, and the evaluation met-rics used in this study are described in Section 2. Results are presentedin Section 3. Finally, discussion and conclusions are provided inSection 4 and Section 5, respectively.

2. Materials and methods

2.1. Datasets

2.1.1. SMOS level 3 TB and soil moisture productsSMOS is a joint ESA, CNES (Centre national d'études spatiales), and

CDTI (the Spanish government agency with responsibility for space)mission that was launched on November 2, 2009 (Kerr et al., 2012).

259A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

The SMOS satellite carries an interferometric radiometer that operatesat L-band, with multiple incidence angles, a spatial resolution of35 km at the center of the field of view, a revisit time of 3 days, and as-cending and descending overpasses at 6:00 AM (local time) and6:00 PM, respectively (Kerr et al., 2001; Kerr et al., 2010). GlobalSMOS Level 3 gridded multi-angular TB and soil moisture (top 0–5 cmsurface layer) products (SMOSL3; version R04 + OPER) are generatedand provided by the CATDS (Centre Aval de Traitement des Données)center in France (Kerr et al., 2013). The SMOSL3 products are deliveredfor both orbits i.e. ascending and descending, projected on a global EASE(Equal Area Scalable Earth) grid (V2) 25 km, by the CATDS, and areavailable online via http://www.catds.fr/. The IFS (Integrated ForecastSystem) soil temperature product from the European Centre for Medi-um-Range Weather Forecasts (ECMWF) is used in the SMOSL3 algo-rithm to retrieve the SMOSL3 soil moisture.

SMOSL3 TB product provides multi-angular TB data (in Kelvin) atthe top of the atmosphere, i.e. not at the surface level and without cor-rection for select reflected extraterrestrial sky (e.g., cosmic and galactic)and atmosphere contributions, but after projection onto the Earth refer-ence frame (unlike the operational Level 2 product). The multi-angularTB are binned and averaged in 5°-width incidence angle bins with thecenter ranging from 2.5° to 62.5°. CATDS has recently providedSMOSL3 TB at 40°, and for this purpose themulti-angular TB are binnedand averaged in 2°-width incidence angle bins. SMOSL3 soil moistureproducts (provided inm3/m3) are derived from themultiangular inver-sion of the L-MEBmodel (L-MEB algorithm) (Wigneron et al., 2007), i.e.the same method used for Level 2 soil moisture retrieval (Kerr et al.,2012), but are improved by using several revisits simultaneously (Kerret al., 2016). SMOS TBs for ascending passes only and their associatedsoil moisture retrievals were used in this study (Al-Yaari et al., 2014a;Al-Yaari et al., 2014b) for a better consistency with SMAP soil moistureretrieval datasets, which are only provided at 6:00 AM.

Radio Frequency Interferences (RFI) originating from man-madeemissions have been shown to affect the quality of the SMOS TB obser-vations (Oliva et al., 2012). RFI probability is used to filter the SMOSdatasets. This probability is the total number of deleted TBs due tosuspected RFI on a certain period divided by the total number of TBmeasurements acquired during the same period available in the SMOSL1C datasets. In this study, SMOSTB and soilmoisture datawere exclud-ed when the RFI probability is higher than 20% following Kerr et al.(2016). The reader is referred to the Algorithm Theoretical Based Docu-ment (Kerr et al., 2013) for more details on the SMOSL3 products.

2.1.2. SMAP level 3 TB and soil moisture productsSMAP is a NASA satellite that was launched on January 31, 2015. The

SMAP satellite at launch carried two instruments: a Synthetic ApertureRadar and a radiometer operating at L-band, with a fixed incidenceangle of 40°, a spatial resolution of 40 km, a revisit of 2–3 days and as-cending and descending overpass at 6:00 PM (local time) and6:00 AM, respectively (Entekhabi et al., 2010; Piepmeier et al., 2016).Soil moisture (top 0–5 cm surface layer) and freeze/thaw were sup-posed to be providedwith three spatial resolutions ~3 km (high-resolu-tion from radar), ~9 km (intermediate-resolution from radar andradiometer), and ~36 km (low-resolution from radiometer), projectedon the EASE V2 grid. However, the radar instrument onboard SMAP sat-ellite stopped transmitting data on Jul 7, 2015 due to a problem in theradar's high-power amplifier (Chan et al., 2016). Currently, soil mois-ture products are retrieved from SMAP TB radiometer data using thebaseline Single Channel Algorithm (SCA) V-pol (Chan et al., 2016;Jackson, 1993). The global daily SMAP Level 3 V3 gridded descendingTB (at both H and V polarizations) and soil moisture (which is a compi-lation of 24 h of L2 soil moisture orbits) products, henceforth referred tohere as SMAP_SCA, on EASE 2 grid (36 km)were used in this study. Un-like the SMOSL3 TBs product, the TBs provided within the SMAP L3product are calibrated at the surface level, i.e. corrected for Sky radiationand atmosphere contributions using auxiliary near surface information

(De Lannoy et al., 2015). The SMAP L2 half-orbit soil moisture product(and also the SMAP L3 soil moisture product) uses ancillary data (in-cluding soil temperature information) from theNASA's GlobalModelingand Assimilation Office (GMAO/GEOS-5) Forward Processing product,which is provided with SMAP datasets. They are freely available fromthe National Snow and Ice Data Center (NSIDC). For more details onthe SMAP mission and SMAP passive products, the reader is referredto (Chan et al., 2016; Piepmeier et al., 2016) and the SMAP Level 2 & 3Soil Moisture (Passive) Algorithm Theoretical Basis Document(SMAP_ATBD) available here: https://nsidc.org/data/docs/daac/smap/sp_l2_smp/pdfs/L2_SM_P_ATBD_v7_Sep2015-po-en.pdf.

Furthermore, it should be noted that both the SMOS and SMAP TBand soil moisture datasets were filtered prior to the regression analysisand the evaluations. A pixel was masked out when:

(i) it is not considered as “Land” in the United States Geological Sur-vey (USGS) Land-Sea mask (water fraction above 10%);

(ii) it is classified as “Urban and Built-Up”, “Snow and Ice”, “Water”,“Permanent Wetlands”, “Evergreen Needleleaf Forest”, or “Ever-green Broadleaf Forest” according to the International GeosphereBiosphere Programme (IGBP) land cover map;

(iii) the number of retrievals b 15 over the whole retrieval period;(iv) it corresponds to a datewhere theMERRA-Land soil temperature

is b274 K (to avoid frost and frozen conditions);(v) SMOS TB is estimated to be not sensitive to the surface effects ac-

cording to the mask developed by Parrens et al. (2016); and(vi) it corresponds to a date that is not recommended for retrieval

based on the SMAP quality flag.

2.1.3. MERRA-Land soil temperatureThe soil temperature product was extracted from the NASAMERRA-

Land product, which is a land-surface model forced with atmosphericreanalysis fields (precipitation corrected using gauges) (Reichle et al.,2011). MERRA_Land is a supplemental land surface data product ofthe Modern-Era Retrospective analysis for Research and Applications(MERRA) datasets, produced by the Goddard Earth Observing Systemmodel and assimilation system. MERRA-Land uses an updated catch-ment land surface model (version Fortuna-2.5) and includes a gauge-based precipitation data from the NOAA Climate Prediction Centre.The accuracy and precision of MERRA-Land soil temperature wereassessed and analyzed by Parinussa et al. (2011) and Holmes et al.(2012). These studies found the performance ofMERRA-Land to be sim-ilar to the ECMWF soil temperature products. TheMERRA-Land productis available for the 1980 - February 2016 period, provided with hightemporal resolution (hourly) and a horizontal resolution of 2/3° longi-tude by 1/2° latitude (http://gmao.gsfc.nasa.gov/research/merra/merra-land.php). The follow-up re-analysis product is MERRA2. It hasimproved soil temperature estimates, and uses gage information to cor-rect the precipitation (Reichle et al., 2016), similarly to (but not exactlythe same as) MERRA-Land. At the time of writing, the MERRA2 productwas not yet available. Consequently, this study usesMERRA-Land auxil-iary information.

2.1.4. ECMWF soil temperatureThe global atmospheric reanalysis ERA-Interim soil temperature

datasets obtained from ECMWF were used in this study. The Hydrolo-gy-Tiled ECMWF Scheme for Surface Exchange over Land (H-TESSEL)is used by the ECMWF forecasts to solve for several parameters includ-ing a four-layer soil temperature profile (Balsamo et al., 2009). In thisstudy, the soil temperature from the first layer (0–0.07 m) provided at00:00, 06:00, 12:00, 18:00 UTC over a grid with a space sampling of0.25 × 0.25° was used. The ECMWF product is available from 1979 topresent. The ECMWF datasets can be freely accessed at: http://apps.ecmwf.int/datasets/ and more information can be found in Berrisfordet al. (2011).

260 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

2.1.5. Ground-based measurementsValidation of remotely sensed soil moisture products against

ground-basedmeasurements is a necessary step before any use. Nowa-days several soil moisture networks share ground-based soil moisturemeasurements via the website of the International Soil Moisture Net-work (ISMN; Dorigo et al., 2011; Dorigo et al., 2015). ISMN is an ESAfunded project initiated through the SMOS CAL/VAL. Data can be freelyobtained from ISMN at https://ismn.geo.tuwien.ac.at/. All sites fromISMN that provide soil moisture within the period of Apr 2015–Jul2016were used in this study to evaluate the remotely sensed soil mois-ture products. Most of the sites are located in different regions with dif-ferent vegetation, climate, and soil conditions.

Eleven networks in North America, Australia, Africa, and Europewere used namely: the PBO_H2O (http://xenon.colorado.edu/portal)network (Larson et al., 2008), the SCAN (Soil Climate Analysis Network)network (http://www.wcc.nrcs.usda.gov/scan/) (Schaefer et al., 2007),the SNOTEL (Snow Telemetry) network (http://www.wcc.nrcs.usda.gov/snow/), the USCRN (U.S. climate reference) network (Bell et al.,2013), the newly built RSMN (Romanian Soil Moisture & TemperatureObservation Network) network (http://assimo.meteoromania.ro/) inRomania, the FMI (Finnish Meteorological Institute) network(Rautiainen et al., 2012) in Finland, the Oznet (Australian MoistureMonitoring Network) network (Smith et al., 2012) in Australia, theSMOSMANIA (Soil Moisture Observing System–Meteorological Auto-matic Network Integrated Application) network (Albergel et al., 2008;Calvet et al., 2007) in France, the DAHRA network (Tagesson et al.,2015) in Senegal, the iRON (Integrated Roaring Fork Observation Net-work) network http://ironagci.blogspot.co.at/, and the REMEDHUS

Fig. 1. IGBP (International Geosphere Biosphere Programme) land cover classification (Friedl etthe USA (c), and Africa and Europe (d).

(Soil Moisture Measurement Stations network of the University of Sala-manca) network (Sanchez et al., 2012) in Spain. To ensure the highquality of the in situmeasurements and to minimize the systematic dif-ferences between them and the remotely-sensed soil moisture prod-ucts, we restricted the validation step to sites with a top soil layer of~0–5 cm and with a number of daily observations N 15. ISMN qualityflags associated with the soil moisture data were applied (Dorigo etal., 2013). Consequently, ~400 (out of ~1000) sites from 11 networkswere used for the evaluation. Moreover, if multiple sensors fall withinone pixel, each sensor is treated independently: in this paper, unlike(De Lannoy&Reichle, 2015)wewill not seek to construct reliable statis-tical confidence intervals. Fig. 1 shows the locations of the different insitu soil moisture sites.

2.2. Methodology

The methodology used in this study to retrieve soil moisture fromSMAP TB based on regression coefficients, obtained from SMOS TB andsoil moisture, is schematized in Fig. 2.

It consists of two steps: the calibration and the data production.

2.2.1. CalibrationA regression equation was analytically derived from the general

(tau-omega) model equations (Mo et al., 1982) by Wigneron et al.(2004):

ln SMð Þ ¼ a2 ln ΓPVð Þ þ a1 ln ΓPHð Þ þ a0 ð1Þ

al., 2010)with the locations of the different in situ soil moisture sites (a) over Australia (b),

Fig. 2. The LRM (local regression method) algorithm: inputs (in orange), calibration step (in blue), and soil moisture production step (in yellow). (For interpretation of the references tocolour in this figure legend, the reader is referred to the web version of this article.)

261A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

where a0, a1, and a2 are regression coefficients, and the first (second)term on the right hand side of Eq. (1) represents the surface reflectivityat vertical (horizontal) polarization (ΓP), described as:

ΓP ¼ 1−TBP

TGð2Þ

where:

TBP is the brightness temperature at polarization p (H or V) at 40° in-cidence angle and TG is the surface soil temperature.We used a multiple linear regression i.e., a statistical technique thatpredicts the outcome of a response (dependent) variable using twoor more independent (explanatory) variables. The coefficients ofEq. (1) were estimated using ordinary least squares techniquesthat minimize the sum of the squared errors. Eq. (1) was used inthis study to retrieve soil moisture from the SMAP L-band TB obser-vations. The coefficients a0, a1, and a2 of Eq. (1) were calibrated perland cover category (obtained from the IGBP land cover map (seeFig. 1)) using themost recent available re-processed SMOS datasets:the SMOS TB in both V and H polarizations at incidence angle of 40°and soil moisture, andMERRA-Land TG datasets. The calibration wasdone during the 2013–2014 period. Note that this calibration wasmade here per land cover category, and not on a pixel to pixelbasis as it was made previously in most LRM studies based onspace-borne observations (Al-Yaari et al., 2016; Parrens et al.,2012; Saleh et al., 2006). This choice was made here to increasethe spatial coverage. SMOS TBs observations are highly affected byRFI and therefore most of the regions in Europe and Asia would bemasked out. To overcome this issue, we removed pixels with highRFIs and then we calibrated the regression equation with the restof pixels within each land cover category. The regression coefficientsare mainly sensitive to the vegetation structure and for a given IGBPvegetation class, the general vegetation structure is similar,

whatever the climate or geographic area. So, we think it is not usefulto distinguish further the vegetation classification depending on thelatitude/climate. Lastly, we applied the obtained coefficients to theSMAP TB data (which are less impacted by RFI) for all the pixelsfor each land cover category.

2.2.2. Soil moisture productionSoil moisture was computed from the SMAP TB data for the Apr

2015 - Jul 2016 period using the regression coefficients computedin the calibration step using Eq. (1). This was done given the factthat we have all inputs for Eq. (1) to compute the soil moisture: TBobservations at both polarizations from SMAP, TG datasets based onthe GMAO GEOS-5 (here in after referred to as GEOS-5) model pro-vided with the SMAP datasets (or any other source like ECMWF),and the coefficients (a0, a1, and a2) from the calibration step.

2.3. Metrics used for evaluating the soil moisture dataset

The SMAP_Reg soil moisture product, obtained using the LRM algo-rithm, was compared with the SMAP and SMOS official Level 3 soilmoisture products, and all three remotely sensed soil moisture productswere evaluated against in situ observations. This was achievedusing classical metrics: Root Mean Square Error (RMSE; m3/m3), Bias(m3/m3), UnbRMSE (m3/m3) (Entekhabi et al., 2010), and the (Pearson)correlation coefficient (R), which can be computed as follows:

Bias ¼ 1N∑N

i¼1Si−Oi ð3Þ

RMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1N∑N

i¼1 Si−Oið Þ2r

ð4Þ

UnbRMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiRMSE2−Bias2

qð5Þ

262 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

R ¼ 1N−1ð Þ∑

N

i¼1

Si−Sσ S

!Oi−OσO

!ð6Þ

where the overbar indicates the mean;

• Si is the ith remotely sensed soil moisture value;• Oi is the ith in situ observed or the remotely sensed soil moisturevalue;

• N is the total number of observations; and• σo and σs are the standard deviations of the in situ observed or re-motely sensed soil moisture values, respectively.

Moreover, Taylor diagrams (Taylor, 2001) were used in this study tocompare in a comprehensiveway the remotely sensed soilmoisture andthe in situ soil moisturemeasurements. Three statistics are summarizedin a Taylor diagram: the normalized standard deviation (SDV) displayedas a radial distance, the correlation coefficient (R) displayed as an anglein the polar plot, and the centered RMSE (displayed as the distance to

Fig. 3.Global inter-comparison between SMOSand SMAPTBs during theApr 2015-Jul 2016 at bof) RMSE, (c & g) Bias, and (d & h) UnbRMSE. Pixels with a number of observations lower than

the point (observed) where R and SDV are equal to one) (Albergelet al., 2012). The performance of the remotely sensed soil moistureproducts is considered closest to ground measurements with theshortest distance to R = 1 and SDV = 1 (Albergel et al., 2012).

3. Results

3.1. SMAP and SMOS inter-comparison

Before presenting the regression analyses and evaluating theSMAP_Reg soil moisture product, it is necessary to evaluate and com-pare the measured (retrieved) TB's (soil moisture) from SMOS andSMAP, which is the key to understand the different results, as it is theonly input that changes between the SMOS and SMAP-based SM re-trieval algorithms. For instance, this could help explain the differencesin Bias, RMSE and correlation obtained between the different soil mois-ture products evaluated in this study over the different networks. Forthis purpose, global maps of R, RMSE, Bias, and UnbRMSE betweenSMAP and SMOS TBs and soil moisture were produced. Fig. 3 shows

th polarizations vertical (V-pol; left) and horizontal (H-pol; right): (a& e) correlation, (b&15 are indicated as blank areas.

Table 1Regression coefficients of Eq. (1) calibrated using SMOS Level 3 TB and soil moisture in

Fig. 4.Global inter-comparison between SMOSand SMAPsoilmoisture retrievals during theApr 2015–Jul 2016: (a) correlation, (b) RMSE, (c) Bias, and (d)UnbRMSE. Pixelswith a numberof observations lower than 15 are indicated as blank areas.

263A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

the global maps between SMOS and SMAP TBs during the Apr 2015 Jul2016 period at both polarizations vertical (V-pol; left) and horizontal(H-pol; right): (a& e) correlation, (b & f) RMSE, (c & g) Bias, and (d&h) UnbRMSE. In general, there is a good agreement between SMOSand SMAP TBs at both V-pol and H-pol particularly in terms of temporaldynamics (R mostly N0.8). The RMSE and UnbRMSE values are lowerbetween SMAP and SMOS at V-pol than between SMAP and SMOS atH-pol over some regions (e.g., region of western North Africa). TheRMSE (UnbRMSE) values range mostly between 2 and 8 (4) K overmost of the globe except over some regions (e.g., Western Europe).SMAP presents cold (warm) Bias with respect to SMOS over most ofthe globe (high latitude regions).

Fig. 4 displays global inter-comparison between the operationalSMOS (L3) and SMAP (SCA) Level 3 soil moisture retrievals during theApr 2015- Jul 2016 period: (a) correlation, (b) RMSE, (c) Bias, and (d)UnbRMSE. The correlations between SMOS and SMAP soil moisture re-trievals (Fig. 4a) are very high (between 0.8 and 1) over Australia, cen-tral Asia and USA, and the Sahel, while moderate correlations are foundover the other regions. SMAP is slightlywetter than SMOS under regionswhere the vegetation density is high as well as on coastlines and overdesert areas (Sahara), whereas SMOS is slightly wetter over India andCentral America, and far north (see Fig. 4c). It can be seen in Fig. 4b &d that higher values of RMSE and UnbRMSE are found in regionswhere the vegetation density is moderate or high than over arid andsemi-arid regions. The retrieved soil moisture data from both SMOSand SMAP seem to agree generally well in terms of UnbRMSE (mostlyb0.05 m3/m3).

2013–2014: a0 represents the intercept, a1 and a2 represent the slope of regression linecorresponding to H-pol and V-pol, respectively.

Land cover class a0 a1 a2

Deciduous needleleaf forest 2.671 1.322 0.937Deciduous broadleaf forest 5.184 2.713 0.889Mixed forest 3.848 2.485 0.492Closed shrublands 0.789 1.068 0.242Open shrublands 0.952 0.864 0.478Woody savannas 3.212 1.903 0.643Savannas 1.821 1.534 0.336Grasslands 0.937 1.032 0.391Croplands 0.815 0.867 0.421Cropland/natural vegetation mosaic 0.874 0.626 0.558Barren or sparsely vegetated 1.049 1.830 0.384

3.2. Regression calibration

The regression coefficients a0 (intercept coefficient), a1 (coefficientfor the H polarized TB), and a2 (coefficient for the V polarized TB) inEq. (1), obtained using SMOSL3 TB (V & H) and soil moisture in the cal-ibration step, are presented in Table 1 and Fig. 5. A unique coefficientvalue for each land cover category was obtained. It can be seen thatthe values of the coefficient for the different land cover categories canbe clearly distinguished. For instance, lowest coefficients values for a0and a2were obtained over “Closed shrublands” andhighest over “Decid-uous needleleaf forest” for a2 and “Deciduous broadleaf forest” for a0

while lowest values for a1 were obtained over “Cropland/Natural vege-tation mosaic” and highest over “Deciduous broadleaf forest”.

In order to make a first evaluation of the quality of the calibrationstep, and before applying the LRM algorithm to the SMAP TB data, weapplied the LRM equations to the SMOS TB data. The soil moisture prod-uct (SMOS_Reg)was retrieved fromSMOSL3 TBusing the regression co-efficients computed from Eq. (1) over the calibration period. Then,SMOS_Reg was compared with the reference SMOSL3 soil moisture forthe same period in terms of RMSE and correlation coefficient (p-value b 0.05).

Looking at the correlation map in Fig. 6a, a remarkable agreement(R N 0.8) can be seen between SMOS_Reg and SMOSL3 over most ofthe globe except over some forests areas (e.g., boreal regions) wherethe correlation values drop below 0.4. Looking at the RMSE map inFig. 6b, the spatial patterns of the RMSE values are also found to be incorrespondence with the vegetation distribution: low RMSE values(~0.05 m3/m3) are found over areas with low vegetation while highRMSE values are found over moderately vegetated regions.

In addition, the regression parameters were applied to the SMOS TBand ECMWF TG dataset as a validation exercise, for a period out of thecalibration period i.e. Apr 2015–Jul 2016. Correlation and RMSE werecomputed between the retrieved SM (using LRM) and SMOSL3

Fig. 5. Regression coefficients of Eq. (1) calibrated using SMOS Level 3 TB and soil moisture in 2013–2014 (a): intercept (a0), (b) slope of regression line corresponding to the horizontalpolarization (a1), and (c) slope of regression line corresponding to the vertical polarization (a2).

264 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

(Fig. 7). The performance of LRMover the validation period is as good aswhat was obtained for the calibration period in terms of both correla-tion and RMSE over most of the globe.

We also added a vegetation index to account for vegetation changesaswas done in Santamaría-Artigas et al. (2016) andMattar et al. (2012).TheNDVI product, obtained fromMODIS (Moderate Resolution ImagingSpectroradiometer), was included in Eq. (1) as follows:

ln SMð Þ ¼ a3NDVIþ a2 ln ΓPVð Þ þ a1 ln ΓPHð Þ þ a0 ð7Þ

The performance of calibration was slightly improved in terms of Rand RMSE particularly over Australia (Fig. 8). However, in this study,

we preferred to keep the regression algorithm as independent as possi-ble of ancillary data (namely the MODIS NDVI dataset) and thereforethis was not taken into account in the subsequent analyses.

3.3. SMAP_Reg soil moisture evaluation

This section presents an evaluation of the SMAP_Reg soil moistureproduct, which was based on applying the LRM algorithm to theSMAP Level 3 TB observations (see Section 2.2) and using two soil tem-perature products: GEOS-5 and ECMWF. Note that the LRM algorithmwas calibrated with MERRA-Land soil temperature; here, we onlychange the input, not the linear regression coefficients, as it will become

Fig. 6. Comparison between the soil moisture values computed from the SMOS TB data using LRM (SMOS_Reg) and the SMOS official Level 3 soil moisture products in 2013–2014: (a)correlation, R and (b) RMSE (m3/m3). Pixels with a number of observations lower than 15 are indicated as blank areas.

265A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

clear that the temperatures estimated inGEOS-5 and ECMWFdonot dif-fer much in the first order. Both these SMAP_Reg soil moisture productswere first compared with the SMAP operational Level 3 soil moistureproduct (SMAP_SCA) to investigate the similarity/dissimilarity betweenthe various SMAP soil moisture products. This was done by computingthe R and RMSE statistical criteria between the SMAP_Reg andSMAP_SCA soil moisture products at the global scale. The temporal cor-relation between SMAP_Reg and SMAP_SCA soil moisture retrievals isshown in Fig. 9a using GEOS-5 and in Fig. 9c using ECMWF. Fig. 9a & cshow that the temporal dynamics of both SMAP_Reg and SMAP_SCAsoil moisture products are generally very similar with R values largerthan 0.8 over most of the globe. However, weaker correlations betweenSMAP_Reg (ECMWF) and SMAP_SCA (GEOS-5) than betweenSMAP_Reg (GEOS-5) and SMAP_SCA (GEOS-5) can be seen over a fewregions particularly over high latitude areas and Sahara. Fig. 9b showsthat the distribution of the RMSE values between SMAP_Reg andSMAP_SCA soil moisture products present clear spatial patterns: lowRMSE values over deserts and savannahs (e.g., the Sahara, Australia,Southern Africa, etc.), whereas high values of RMSE values were gener-ally found over vegetated areas. Looking at both Fig. 9a & c (correla-tions) and Fig. 9b & d (RMSE), there is a general good agreementbetween SMAP_Reg and SMAP_SCA over regions with low to moderateamounts of vegetation cover.

The SMAP_Reg soil moisture product with GEOS-5 soil temperatureas auxiliary input was additionally evaluated against

i) in situ soil moisture observations using N400 sites from eleven net-works spread over the globe (see Section 2.1.5); and

ii) the operational SMAP and SMOS Level 3 soil moisture products atthese individual site locations. The SMAP_SCA and the SMOSL3 soil

moisture products were considered in the evaluation in order to in-vestigate the consistency in time variations between the new soilmoisture product (SMAP_Reg) and the original ones at differentsites.

Taylor diagrams (see Section 2.3) given in Fig. 10 show the statisticsfor the sites individually. Fig. 10 shows values of SDV, R, and the cen-tered RMSE between the remotely sensed soil moisture products andmeasured soil moisture values over all sites used in this study. In Fig.10, the performance of SMAP_SCA (blue symbols), SMAP_Reg (red sym-bols), and SMOSL3 (green symbols) varies from one site to another andfrom one network to another as demonstrated by the uneven distribu-tion of the sites (shown as circles) in the plots. Looking at the Taylor di-agram over the SCAN and SNOTEL sites, the correlations values rangebetween0 and 0.9 and both SMAP_Reg and SMOSL3 tend to have higherSDV values than SMAP_SCA. Over the REMEDHUS sites, the three algo-rithms have a comparable performance but with a large variability (asdefined byhigh SDV index)with the in situ observations; the correlationvalues range between 0.4 and 0.8. Over the PBO-H2O sites, SMAP_SCApatterns are closer to the in situ patterns than SMOSL3 and SMAP_Reg,which have larger SDV values than the in situ observations and mostof the correlation values range between 0.5 and 0.9. Over the USCRNsites, the retrieved soil moisture values from all algorithms have a sim-ilar variability to that of the in situ observations, although SMAP_Regand SMOSL3 soil moisture products have a larger variability for somesites. Over the RSMN sites, the three products present the same levelof performance with respect to the in situ observations but all hadhigher SDV than the in situ observations. However, the correlationsdrop below 0.4 for a few sites for the three products. Over the DAHRA

Fig. 7. Comparison between SMOS-Reg (retrieved based on regression equation) and SMOSL3 (retrieved based on the L-MEB model) in terms of (a) correlation and (b) RMSE (m3/m3)during the Apr 2015–Jul 2016 period. Pixels with a number of observations lower than 15 are indicated as blank areas.

266 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

network, the SMOSL3 (best correlation) and SMAP_SCAproduct lie clos-est to the observed point followed by SMAP_Reg with higher SDV thanthe in situ observations for all products. Over the iRON sites, the correla-tions range from 0.4 to 0.8 for the SMOSL3 and SMAP_Reg. SMAP_SCAand SMAP_Reg are comparable in terms of variability and they are clos-er to the in situ patterns than SMOSL3. Over the Oznet network, thethree products are comparable in terms of correlations (ranging from0.6 to 0.9) and, similarly to the RSMN network, overestimate the insitu observations. Over the FMI sites, the correlations range from 0.4 to0.8 and 0.9 for SMAP_SCA and SMAP_Reg, respectively. Similarly, towhat was obtained over the Oznet sites, the SMAP_SCA and SMAP_Regproducts overestimate the in situ observations overmost of the sites. Fi-nally, over SMOSMANIA, SMAP_Reg has better correlations, with in situdata ranging from 0.4 to 0.95, than the other two products. For SMOSL3,the correlationsdrop below0.4 for a few sites. Overall, all three productshave approximately the same level of performance in terms of variabil-ity but SMAP_SCA is slightly better in terms of temporal dynamics.

In order to have a general idea on the overall performance ofSMAP_SCA, SMAP_Reg, and SMOSL3, the average of the Bias, RMSE,and UnbRMSE values and the median of the correlation values for allsites were computed per each network. Given the different sizes of thenetworks, the varying sample sizes, and varying temporal and spatialautocorrelations, the values are only indicative and no statistical confi-dence levels are provided. It should be noted that the average and me-dian values were only computed when the site has a number ofobservations N 15 and p-value b 0.05. This is presented in Table 2 andFig. 11, which show the comparison statistics for the three algorithmsi.e. SMAP_Reg, SMAP_SCA, and SMOSL3 against the in situ observations.It can be seen in Table 2 and Fig. 11 that the best R scores (R N 0.80) for

the three algorithmswere obtained over the Oznet siteswhile theworstones were observed (R ~ 0.58) over the SNOTEL sites for SMAP_SCA andover the DAHRA and SMOSMANIA sites (R = 0.51 and R = 0.45 forSMAP_Reg and SMOSL3 respectively). SMOSL3 had highest R valuesover the DAHRA site while SMAP_Reg had highest R values overSMOSMANIA. Other than those two networks, SMAP_SCA had highestR values. However, both SMAP products i.e. SMAP_SCA and SMAP_Reghave comparable performance particularly in terms of correlation coef-ficients. In terms of UnbRMSE, SMAP_SCA had lower values for all sitesexcept in Oznet where the lowest values were obtained by SMOSL3.Even though it is difficult to compare absolute values at in situ locations,a comparison based on a large sample can give some indication ofbiases: the Bias values showed that all products are generally dry, ex-cept over RSMN, DAHRA, and Oznet (and FMI for SMAP_SCA). UnlikeSMAP_Reg, a notable overall positive Bias is obtained over FMI forSMAP_SCA (overestimation). The overestimation of in situ soil moistureobservations over RSMN and FMI networks by SMAP_SCA is in linewiththe recent findings of Zeng et al. (2016).

4. Discussion

We investigated the potential utility of a physically based multi-lin-ear regression approach to retrieve soil moisture from two microwaveremote sensing satellites that operate at L-band: SMOS and SMAP. Theapproach consists of two steps:

(i) a calibration step to compute regression coefficients using SMOSTB and soil moisture over 2013–2014 (calibration period); and

(ii) a production step to retrieve soil moisture from SMAP TB using

Fig. 8. Comparison between SMOS_Reg soil moisture values (retrieved based on regression equation with including NDVI datasets) and SMOSL3 (retrieved based on the L-MEB model)during the 2013–2014 period in terms of: (a) correlation and (b) RMSE (m3/m3). Pixels with a number of observations lower than 15 are indicated as blank areas.

267A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

the computed regression coefficients, for Apr 2015 to Jul 2016(production period).

4.1. SMAP and SMOS inter-comparison

Before applying the regressions, an inter-comparison was made be-tween SMOS and SMAP TBs and soilmoisture. From the results (Figs. 3 &

Fig. 9. Comparison between the SMAP-derived soil moisture applying LRM (SMAP_Reg) usingLevel 3 soil moisture product (SMAP_SCA) from Apr 2015 to Jul 2016: (top) correlation, R (−)

4), it was shown that there is a very good agreement between the twodatasets. However, small discrepancies (2 to 4 K of Bias) between theSMOS and SMAP TBs were found over most of the globe and high dis-crepancies were found particularly over regions affected by RFI (West-ern Europe, North Africa, etc.). This is not unexpected due to the factthat, as indicated in Sections 2.1.1 and 2.1.2, the SMOSL3 TB productprovides TBs on top of the atmosphere and there is no correction forsky and atmosphere contributions whereas SMAP TB is provided atthe surface. According to De Lannoy et al. (2015), a difference of b2 K

two different soil temperature products: GEOS-5 (left) and ECMWF (right) and the SMAPand (bottom) RMSE (m3/m3).

Fig. 10. Taylor's diagrams for SMAP_Reg (in red), SMAP_SCA (in blue), and SMOSL3 (in green) over the FMI, DAHRA, iRON, andOznet (upper panel), over SCAN, PBO-H2O, REMEDHUS, andRSMN sites (middle panel), and over SNOTEL, USCRN, and SMOSMANIA sites (lower panel). (For interpretation of the references to colour in thisfigure legend, the reader is referred to theweb version of this article.)

268 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

for H-Pol TB and 1 K for V-Pol TB at 40° incidence angle between SMOSTB and SMAP TB can be attributed to the contributions of the atmo-spheric and reflected sky (e.g., cosmic and galactic) radiations. Thislowdifference can be explained by the fact that the effects of (i) the con-tribution of the atmosphere to TB (direct and reflected) and (ii) the at-tenuation effects due to the atmosphere, partially offset each other.Nevertheless, local and short-term values regularly exceed 5 K (DeLannoy et al., 2015). In addition, SMAP TBs are water-body correctedwhile SMOS TBs are not. Thus, the SMOS and SMAP TBs do not corre-spond exactly to the same pixel coverage: the SMOS TBs correspondto the whole pixel, while the SMAP TBs correspond to the whole pixel,but excluding open water areas. As the emission of open water surfaces(~60–150 k) is small in comparison to the emission of soil and vegeta-tion-covered surfaces, applying this correction leads systematically toan increase in the TBs values. Note that for pixels with a fraction ofwater bodies higher than 10%, data were filtered out. But even afterthis filtering, the water TB correction may have an impact on the TBvalues. This water TB correction may explain the warm “Bias” (~10 K)of the SMAP TBswith respect to the SMOSTBs over high latitude and bo-real regions (where many pixels may contain lakes, rivers, wetlands,etc.). Excluding these regions, a small cold Bias of SMAP TBswith respectto SMOS TBs can be noted (~3–6 K). The Bias between the SMOS and

SMAP TB values might have an impact in our approach, as the calibra-tion step was based on the SMOS TB data, while the regression coeffi-cients were applied to the SMAP TB data. However, no correction wasapplied in the framework of this study, as this Bias is not uniform glob-ally. Overall, a very good agreement was found globally between theSMAP and SMOS TBs data.

4.2. Regression calibration and soil moisture production

The regression model was run for each land cover category (definedhere using the IGBP land cover map) separately and thus we obtainedcoefficients of each land cover category. These values vary from oneland cover category to another, reflecting the different characteristicsfor each land cover category. These three parameters as indicated bySaleh et al. (2006) are a function of the soil type and roughness. The re-gression approach quality was evaluated in two ways:

(i) first, we estimated soil moisture from SMOS TB and comparedthe predicted soil moisture (SMOS_Reg) to the reference(SMOSL3) using correlation and RMSE, over the calibration peri-od (2013–2014). High correlations (R N 0.8) and low RMSEvalues were obtained between SMOS_Reg and the reference

Table 2Statistics of the evaluation of the SMAP_Reg, SMAP_SCA, and SMOSL3 against ground basedmeasurements. The average (median for R instead of “average” as correlation is not additive)values were only computed for sites that have p_value b 0.05 and number of observations was higher than 15.

Matric Bias (m3/m3)(mean)

RMSE (m3/m3)(mean)

R (median) UnbRMSE (m3/m3)(mean)

Algorithm

Network(No of sites)

SMAP_SCA SMAP_Reg SMOSL3

SMAP_SCA SMAP_RegSMOS

L3SMAP_SCA SMAP_Reg

SMOSL3

SMAP_SCA SMAP_RegSMOS

L3

REMEDHUS (19)

–0.021 0.007 –0.047 0.087 0.094 0.093 0.65 0.63 0.61 0.051 0.060 0.052

PBO_H2O(76)

–0.024 –0.008 –0.022 0.055 0.056 0.067 0.75 0.76 0.68 0.044 0.049 0.056

RSMN(16)

0.054 0.044 0.026 0.082 0.081 0.086 0.67 0.64 0.54 0.051 0.059 0.069

SCAN(104)

–0.022 –0.013 –0.026 0.075 0.085 0.090 0.73 0.70 0.60 0.050 0.058 0.063

SNOTEL(125)

–0.053 –0.052 –0.047 0.096 0.099 0.103 0.58 0.54 0.55 0.063 0.066 0.071

USCRN(51)

–0.027 –0.017 –0.034 0.075 0.081 0.091 0.70 0.71 0.61 0.046 0.054 0.059

FMI(14)

0.076 –0.032 – 0.109 0.085 – 0.61 0.59 – 0.029 0.035 –

iRON(3)

–0.094 –0.147 –0.182 0.104 0.153 0.190 0.64 0.54 0.50 0.036 0.039 0.051

DAHRA(1)

0.017 0.066 0.007 0.054 0.108 0.050 0.64 0.51 0.82 0.051 0.085 0.050

SMOSMANIA(7)

–0.070 –0.095 –0.097 0.110 0.123 0.127 0.63 0.69 0.45 0.045 0.042 0.062

Oznet(34)

0.016 0.040 0.005 0.091 0.112 0.080 0.85 0.84 0.86 0.074 0.089 0.064

269A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

over the continental surfaces, indicating that the spatio-temporaldynamics of SMOSL3 were well captured by SMOS_Reg. Howev-er, some differences can be noted in terms of magnitude overhigh to moderate vegetation areas and high latitude regions.This is not unexpected due to uncertainties in the SMOS datasetscaused by the high vegetation attenuation effects in these re-gions, which is a major problem for most of the remotely sensedsoil moisture retrievals (Vittucci et al., 2016; Wigneron et al.,2003); and

(ii) Second, we estimated soil moisture from SMAP TB for the Apr2015 to Jul 2016 period (SMAP-Reg) and compared it to the op-erational SMAP_SCA and the SMOSL3 soil moisture productsagainst N400 sites over the world.

Soil temperature is an important input in the radiative transfer equa-tion and has a significant impact on the final estimate of the soil mois-ture retrievals (Holmes et al., 2012; Lv et al., 2016; Parinussa et al.,2011). In order to study how sensitive is the retrieved soil moisture tothe soil temperature effects, we used soil temperature from two differ-ent sources: ECMWF and GEOS-5. We applied the regression coeffi-cients to SMAP TB using these two products and then we comparedwith SMAP_SCA. It was found that in general the spatial patterns aresimilar for both products in terms of R and RMSE values; however, thecorrelations between SMAP_Reg (ECMWF) and SMAP_SCA (GEOS-5)are lower than the correlations between SMAP_Reg (GEOS-5) andSMAP_SCA (GEOS-5) over some regions (e.g., Sahara, Far Eastern Feder-al District, East-Central Canada, etc.). The better agreement betweenSMAP_Reg (GEOS-5) and SMAP_SCA (GEOS-5) does not necessarilymean that the quality of GEOS-5 is better than ECMWF. This could sim-ply results from the fact that the same soil temperature product wasused in both the regression approach and the SMAP_SCA algorithm.However, it does seem that soil temperature has an important impacton the soil moisture retrievals (e.g., Lv et al., 2016), i.e. using the same

soil temperature leads to similar soil moisture retrievals from SMAPnomatter if different retrieval approaches are used. This could partiallyexplain the strong agreement found between SMAP_SCA andSMAP_Reg.

Results from the comparison between SMAP_Reg both (ECMWF &GEOS-5) and SMAP_SCA showed that SMAP_Reg is in agreement withSMAP_SCA, particularly in terms of temporal dynamic which is of highrelevance (Crow et al., 2010; Liu et al., 2012). Moreover, it is noticedhere that the performance of the “production” step i.e. comparison be-tween SMAP_Reg and SMAP_SCA is much better than the “calibration”step i.e. comparison between SMOS_Reg and SMOSL3. This can be partlyexplained by three reasons:

1- in the calibration step, the TBs used are not exactly the samewhile inthe production step they are. More specifically, the SMOSL3 soilmoisture product is not directly retrieved from SMOSL3 TB butfrom TB products in the Fourier domain (L1B); thus the TB used inthe regression does not necessarily match the actual TB used to re-trieve SMOSL3 soilmoisture. However, it is expected that the tempo-ral dynamics of the two soil moisture products will be more similarbecause they will be driven by the common input TB dynamics.This uncertainty, among others, may affect the quality of thecalibration;

2- the quality of SMAP TB seems to be very good and therefore whatev-er the used algorithm, the resultant soil moisture is the same partic-ularly in terms of temporal dynamics. This, again, may partiallyexplain the strong similarity between the two products i.e.SMAP_Reg and SMAP_SCA; and

3- the regression (LRM coefficients) is based onMERRA-Land TG, whichis similar to the GEOS-5 product used in SMAP_SCA and not inSMOSL3 (for which the ECMWF product is used); explicitly indicat-ing that soil temperature may play a crucial role in the quality of theSM retrievals.

Fig. 11. Bar charts showing Bias (m3/m3), R, RMSE (m3/m3), and UnbRMSE (m3/m3) between SMAP_SCA (in blue), SMAP_Reg (in red), and SMOSL3 (in green) and the observed soilmoisture over the 11 networks used in this study. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

270 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

4.3. SMAP_Reg soil moisture evaluation

Results from evaluating both SMAP products and SMOSL3 against insitu observations showed that SMAP_SCA (slightly better) andSMAP_Reg soil moisture products have comparable performance withsimilar R values over the REMEDHUS, PBO_H2O, SNOTEL, SCAN, Oznet,and USCRN sites. Nevertheless, all three remotely sensed soil moistureproducts had poor performance over the SNOTEL sites. The poor perfor-mance of the three algorithms over the SNOTEL network can be attrib-uted to several reasons: among them it should be considered thatmost of the SNOTEL sites are located in mountain regions with forestsand freezing and thawing processes, more details on these aspects canbe found in Al Bitar et al. (2012). It should be kept in mind that whenusing only the recommended retrievals for SMAP, there was no dataleft from the iRON sites. So the statistics for this particular network

were based on data without considering if the retrieval was recom-mended or not but other filters were, however, applied.

All remotely sensed soil moisture products underestimated general-ly the in situ observations used in this study. The Bias values rangedfrom −0.147 m3/m3 (iRON) to 0.066 m3/m3 (DAHRA) for SMAP_Reg,from −0.094 m3/m3 (iRON) to 0.076 m3/m3 (FMI) for SMAP_SCA, andfrom −0.182 m3/m3 (iRON) to 0.026 m3/m3 (RSMN) for SMOSL3. Thisso-called “dry” Bias of SMOSL3 and SMAP_SCA is in line with previousstudies (Al-Yaari et al., 2014a; Al Bitar et al., 2012; Chan et al., 2016;Dente et al., 2012). On the other hand, an overestimation was foundover FMI (for only SMAP_SCA) and RSMN, DAHRA, and Oznet sites(for the three products). It should be kept inmind that FMI is a very spe-cific network: the sites of FMI are located in high latitude regions withcold climate in which remotely-send soil moisture retrievals are influ-enced by the effects of soil freezing and thawing processes, organic

271A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

matter in the soil substrate and the presence of numerous water bodiesand bogs (Rautiainen et al., 2012; Zeng et al., 2016). The reasoning be-hind the underestimation/overestimation of the in situ soil moisturevalues is a challenge. The dry Bias could be related to the different spa-tial scales and sampling depths between the satellites and the in situ ob-servations (e.g., Dorigo et al., 2015; Escorihuela et al., 2010; Rondinelli etal., 2015). Moreover, although RFI was severely filtered from the SMOSdatasets, it could be that some sources of RFI are still not filtered/detect-ed, which can explain the rather general underestimation which isfound in this study (Oliva et al., 2016). This was evident especially forSMOS as seen in Fig. 3where SMOS still has higher TBs close to those re-gions despite filtering RFI. The reader is referred to Al Bitar et al. (2012)for a discussion on these questions.

In terms of UnbRMSE, a comparable performance betweenSMAP_SCA and SMAP_Reg was found over the SMOSMANIA, iRON,and SNOTEL sites but lower values were obtained with SMAP_SCAover the other networks. SMOSL3had generally higher UnbRMSE valuesthan both SMAP_SCA and SMAP_Reg. However, SMOSL3 had lowervalues over the Oznet and DAHRA sites.

It was noted from Table 2 and Fig. 11, that the SMAP TB-based soilmoisture products (SMAP_SCA and SMAP_Reg) have a slightly betterperformance than SMOSL3 for most of the networks especially interms of temporal dynamics. Although the three algorithms use TB atL-band and rely on the same radiative transfer equation (tau-omegamodel), they vary inmany things (e.g., ancillary datasets, model param-eterizations and assumptions). For instance, SMOS and SMAP use twodifferent land cover maps: ECOCLIMAP 2004 (containing 213 classes)for SMOS andMODIS IGBP (containing 17 classes) for SMAP, so it is like-ly there is a mismatch between the real land cover and the theoreticalland cover used in SMAP and SMOS soil moisture retrievals leading toa different behavior of the soil moisture retrievals. Thismay also partial-ly explain the similarity between SMAP_Reg and SMAP_SCA given theuse of the same land cover. These differences were already noted inthe direct inter-comparison between SMOSL3 and SMAP_SCA displayedin Fig. 4. The SMOS team is currently investigating the impact of landcover mapping and the possibility to replace the ECOCLIMAP map bythe IGBP map in the SMOS soil moisture retrieval algorithm. Further-more, the better performance of SMAP soil moisture products could berelated to the enhanced quality of the SMAP TB observations due to animproved RFI mitigation and detection system (Piepmeier et al.,2016). Moreover, SMOS TB observations have a radiometric error of~3 to 3.5 K while SMAP TB observations have a radiometric error of~1 K (De Lannoy et al., 2015). Finally, SMOS and SMAP use different sur-face soil temperature sources for their operational products.

Based on the presented results, it can be noted that applying regres-sion analysis to TB (from SMAP and SMOS) observed at L-band(1.4 GHz) gave better results compared to what was found by Al-Yaariet al. (2016), who applied the LRM algorithm to TB observed at C-band (from AMSR-E; 6.9 GHz). This is not unexpected, as the simplifica-tions and assumptions (e.g., neglecting the scattering effects) of the LRMmethod are more valid at L-band. Moreover, both SMAP and SMOS ob-serve TB at the same frequency i.e. L-band, which is considered optimalfor soil moisture retrievals (Chan et al., 2016; Jackson, 1993). On theother hand, a similar behavior with Al-Yaari et al. (2016) of the regres-sion coefficients that correspond to the H polarization and V polariza-tion was found: low (high) coefficient values at H polarizationcorrespond generally to high (low) values at V polarization over mostof the regions.

5. Conclusions

This study demonstrated the potential benefits of combining SMOSand SMAPdatasets given the good performance of SMAP_Reg comparedto SMOSL3 and SMAP_SCA operational products over some regions. Thisin return shows the close similarity between SMOS and SMAP TB obser-vations andhighlights that an integration of SMAPand SMOSdatasets to

build a long term soilmoisture record (with higher temporal frequency)will be successful. Finally, this first evaluation of preliminary SMAPproducts, and the inter-comparison with SMOS datasets provided in-sights and statistics that can be useful for SMAP/SMOS soil moistureproduct validation and SMAP/SMOS algorithm refinements and conver-gence on auxiliary datasets. A calibration of the soil and vegetation ef-fects has been recently made in the SMOS soil moisture retrievalalgorithm (Fernández Morán et al., 2016). A new SMOS soil moistureproduct integrating this new calibration and with a significantly im-proved accuracy is being produced. Future research will consider thecalibration of SMAP-Regwith this new SMOS product and further fusionstudies will be continued by applying LRM to the SMOS and SMAPdatasets considering other important variables such as the vegetationopacity.

Acknowledgment

The authors would like to thank the European Space Agency (ESA;SMOS Expert Support Laboratory) and the CNES SMOS TOSCA (TerreOcéan Surfaces Continentales et Atmosphère) program for funding thisstudy. The authors acknowledge CATDS and the Jet Propulsion Labora-tory and NSIDC for the SMOS and SMAP products, respectively. Wethank the Editor and three anonymous reviewers for their constructivecomments.

References

Aires, F., Aznay, O., Prigent, C., Paul, M., Bernardo, F., 2012. Synergistic multi-wavelengthremote sensing versus a posteriori combination of retrieved products: applicationfor the retrieval of atmospheric profiles using MetOp-A. J. Geophys. Res. Atmos. 117.

Al-Yaari, A., Wigneron, J.P., Ducharne, A., Kerr, Y., de Rosnay, P., de Jeu, R., Govind, A., AlBitar, A., Albergel, C., Muñoz-Sabater, J., Richaume, P., Mialon, A., 2014a. Global-scale evaluation of two satellite-based passive microwave soil moisture datasets(SMOS and AMSR-E) with respect to Land Data Assimilation System estimates. Re-mote Sens. Environ. 149, 181–195.

Al-Yaari, A., Wigneron, J.P., Ducharne, A., Kerr, Y.H., Wagner, W., De Lannoy, G., Reichle, R.,Al Bitar, A., Dorigo, W., Richaume, P., Mialon, A., 2014b. Global-scale comparison ofpassive (SMOS) and active (ASCAT) satellite based microwave soil moisture re-trievals with soil moisture simulations (MERRA-Land). Remote Sens. Environ. 152,614–626.

Al-Yaari, A., Wigneron, J.P., Kerr, Y., de Jeu, R., Rodriguez-Fernandez, N., van der Schalie, R.,Al Bitar, A., Mialon, A., Richaume, P., Dolman, A., Ducharne, A., 2016. Testing regres-sion equations to derive long-term global soil moisture datasets from passive micro-wave observations. Remote Sens. Environ. 180, 453–464.

Al Bitar, A., Leroux, D., Kerr, Y.H., Merlin, O., Richaume, P., Sahoo, A., Wood, E.F., 2012.Evaluation of SMOS soil moisture products over continental U.S. using the SCAN/SNOTEL network. IEEE Trans. Geosci. Remote Sens. 50, 1572–1586.

Albergel, C., Rüdiger, C., Pellarin, T., Calvet, J.-C., Fritz, N., Froissard, F., Suquia, D., Petitpa,A., Piguet, B., Martin, E., 2008. From near-surface to root-zone soil moisture usingan exponential filter: an assessment of the method based on in-situ observationsand model simulations. Hydrol. Earth Syst. Sci. Discuss. 12, 1323–1337.

Albergel, C., de Rosnay, P., Balsamo, G., Isaksen, L., Muñoz-Sabater, J., 2012. Soil moistureanalyses at ECMWF: evaluation using global ground-based in situ observations.J. Hydrometeorol. 13, 1442–1460.

Balsamo, G., Viterbo, P., Beljaars, A.C.M., van den Hurk, B.J.J.M., Hirschi, M., Betts, A.K., K., S.,2009. A revised hydrology for the ECMWF model: verification from field site to ter-restrial water storage and impact in the ECMWF-IFS. J. Hydrometeorol. 10.

Bartalis, Z., Hasenauer, S., Naeimi, V., Wagner,W., 2007.WARP-NRT 2.0 ReferenceManual.ASCAT Soil Moisture Report Series, No. 14Institute of Photogrammetry and RemoteSensing, Vienna University of Technology, Austria.

Bell, J.E., Palecki, M.A., Baker, C.B., Collins, W.G., Lawrimore, J.H., Leeper, R.D., Hall, M.E.,Kochendorfer, J., Meyers, T.P., Wilson, T., Diamond, H.J., 2013. U.S. climate referencenetwork soil moisture and temperature observations. J. Hydrometeorol. 14, 977–988.

Berg, A., Lintner, B.R., Findell, K.L., Malyshev, S., Loikith, P.C., Gentine, P., 2014. Impact ofsoil moisture–atmosphere interactions on surface temperature distribution. J. Clim.27, 7976–7993.

Berrisford, P., Dee, D., Poli, P., Brugge, R., Fielding, K., Fuentes, M., Kallberg, P., Kobayashi,S., Uppala, S., Simmons, A., 2011. The ERA-Interim archive Version 2.0, ERA Report Se-ries 1. ECMWF, Shinfield Park. Reading, UK, p. 13177.

Calvet, J.-C., Fritz, N., Froissard, F., Suquia, D., Petitpa, A., Piguet, B., 2007. In situ soil mois-ture observations for the CAL/VAL of SMOS: the SMOSMANIA network. Geoscienceand Remote Sensing Symposium, 2007. IGARSS 2007. IEEE International. IEEE,pp. 1196–1199.

Chan, S., Bindlish, Rajat, O'Neill, Peggy, Njoku, Eni, Jackson, Tom, Colliander, Andreas,Chen, Fan, Burgin, Mariko, Dunbar, Scott, Piepmeier, Jeffrey, Yueh, Simon,Entekhabi, Dara, Cosh, Michael H., Caldwell, Todd, Walker, Jeffrey, Wu, Xiaoling,Berg, Aaron, Rowlandson, Tracy, Pacheco, Anna, McNairn, Heather, Thibeault, Marc,Martínez-Fernández, José, González-Zamora, Ángel, Seyfried, Mark, Bosch, David,

272 A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

Starks, Patrick, Goodrich, David, Prueger, John, Palecki, Michael, Small, Eric E., Zreda,Marek, Calvet, Jean-Christophe, Crow, Wade T., Kerr, Y., 2016. Assessment of theSMAP passive soil moisture product. IEEE Trans. Geosci. Remote Sens. 54, 4994–5007.

Chen, T., McVicar, R.T., Wang, G., Chen, X., de Jeu, A.R., Liu, Y.Y., Shen, H., Zhang, F.,Dolman, J.A., 2016. Advantages of using microwave satellite soil moisture overgridded precipitation products and land surface model output in assessing regionalvegetation water availability and growth dynamics for a lateral inflow receiving land-scape. Remote Sens. 8.

Choudhury, B.J., Schmugge, T.J., Chang, A., Newton, R.W., 1979. Effect of surface roughnesson the microwave emission from soils. J. Geophys. Res. Oceans 84, 5699–5706.

Crow,W.T., Miralles, D.G., Cosh, M.H., 2010. A quasi-global evaluation system for satellite-based surface soil moisture retrievals. IEEE Trans. Geosci. Remote Sens. 48,2516–2527.

De Lannoy, G.J.M., Reichle, R.H., Peng, J., Kerr, Y., Castro, R., Kim, E.J., Liu, Q., 2015.Converting between SMOS and SMAP level-1 brightness temperature observationsover Nonfrozen land. IEEE Geosci. Remote Sens. Lett. 12, 1908–1912.

De Lannoy, G.J.M., Reichle, R.H., 2015. Global assimilation of multiangle andmultipolarization SMOS brightness temperature observations into the GEOS-5 catch-ment land surface model for soil moisture estimation. J. Hydrometeorol. 17, 669–691.

Dente, L., Su, Z., Wen, J., 2012. Validation of SMOS soil moisture products over the Maquand Twente regions. Sensors 12, 9965–9986.

Dorigo, W.A., Wagner, W., Hohensinn, R., Hahn, S., Paulik, C., Xaver, A., Gruber, A., Drusch,M., Mecklenburg, S., van Oevelen, P., Robock, A., Jackson, T., 2011. The internationalsoil moisture network: a data hosting facility for global in situ soil moisture measure-ments. Hydrol. Earth Syst. Sci. 15, 1675–1698.

Dorigo, W.A., Xaver, A., Vreugdenhil, M., Gruber, A., Hegyiová, A., Sanchis-Dufau, A.D.,Zamojski, D., Cordes, C., Wagner, W., Drusch, M., 2013. Global automated quality con-trol of in situ soil moisture data from the international soil moisture network. VadoseZone J. 12.

Dorigo, W.A., Gruber, A., De Jeu, R.A.M., Wagner, W., Stacke, T., Loew, A., Albergel, C.,Brocca, L., Chung, D., Parinussa, R.M., Kidd, R., 2015. Evaluation of the ESA CCI soilmoisture product using ground-based observations. Remote Sens. Environ. 162,380–395.

Enenkel, M., Reimer, C., Dorigo, W., Wagner, W., Pfeil, I., Parinussa, R., De Jeu, R., 2015.Combining satellite observations to develop a daily global soil moisture product fora wide range of applications. Hydrol. Earth Syst. Sci. Discuss. 2015, 11549–11589.

Entekhabi, D., Njoku, E.G., O'Neill, P.E., Kellogg, K.H., Crow,W.T., Edelstein,W.N., Entin, J.K.,Goodman, S.D., Jackson, T.J., Johnson, J., Kimball, J., Piepmeier, J.R., Koster, R.D., Martin,N., McDonald, K.C., Moghaddam, M., Moran, S., Reichle, R., Shi, J.C., Spencer, M.W.,Thurman, S.W., Leung, T., Van Zyl, J., 2010. The soil moisture active passive (SMAP)mission. Proc. IEEE 98, 704–716.

Escorihuela, M.J., Chanzy, A., Wigneron, J.P., Kerr, Y.H., 2010. Effective soil moisture sam-pling depth of L-band radiometry: a case study. Remote Sens. Environ. 114,995–1001.

FernándezMorán, R., Wigneron, J.-P., De Lannoy, G., Lopez-Baeza, E., Mialon, A., Mamoodi,A., Parrens, M., Al Bitar, A., Richaume, P., Kerr, Y., 2016. Calibrating the effective scat-tering albedo in the SMOS algorithm: some first results. 2016 IEEE International Geo-science and Remote Sensing Symposium (IGARSS), Beijing, 2016.

Friedl, M.A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A., Huang, X.,2010. MODIS collection 5 global land cover: algorithm refinements and characteriza-tion of new datasets. Remote Sens. Environ. 114, 168–182.

GCOS, 2010. Implementation Plan for the Global Observing System for Climate in Supportof the UNFCC (2010 Update).

Grant, J.P., Saleh-Contell, K., Wigneron, J.P., Guglielmetti, M., Kerr, Y.H., Schwank, M., Skou,N., Griend, A.A.V.D., 2008. Calibration of the L-MEB model over a coniferous and a de-ciduous forest. IEEE Trans. Geosci. Remote Sens. 46, 808–818.

Hirschi, M., Mueller, B., Dorigo, W., Seneviratne, S.I., 2014. Using remotely sensed soilmoisture for land–atmosphere coupling diagnostics: the role of surface vs. root-zone soil moisture variability. Remote Sens. Environ. 154, 246–252.

Holmes, T.R.H., de Rosnay, P., de Jeu, R., Wigneron, R.J.P., Kerr, Y., Calvet, J.C., Escorihuela,M.J., Saleh, K., Lemaître, F., 2006. A new parameterization of the effective temperaturefor L band radiometry. Geophys. Res. Lett. 33.

Holmes, T.R.H., Jackson, T.J., Reichle, R.H., Basara, J.B., 2012. An assessment of surface soiltemperature products from numerical weather prediction models using ground-based measurements. Water Resour. Res. 48.

Hupet, F., Vanclooster, M., 2002. Intraseasonal dynamics of soil moisture variability withina small agricultural maize cropped field. J. Hydrol. 261, 86–101.

Jackson, T.J., Schmugge, T.J., 1991. Vegetation effects on the microwave emission of soils.Remote Sens. Environ. 36, 203–212.

Jackson, T.J., 1993. Measuring surface soil moisture using passivemicrowave remote sens-ing. Hydrol. Process. 7, 139–152.

Jackson, T.J., O'Neill, P., Njoku, E.S.C., Bindlish, R., Colliander, A., Chen, F., Burgin, M.,Dunbar, S., Piepmeier, J., Cosh, M., Caldwell, T., Walker, J., Wu, X., Berg, A.,Rowlandson, T., Pacheco, A., McNairn, H., Thibeault, M., Martínez-Fernández, J.,González-Zamora, Á., Seyfried, M., Bosch, D., Starks, P., Goodrich, D., Prueger, J., Su,Z., van der Velde, R., Asanuma, J., Palecki, M., Small, E., Zreda, M., Calvet, J., Crow,W., Kerr, Y., Yueh, S., Entekhabi, D., 2016. Soil moisture active passive (SMAP) projectcalibration and validation for the L2/3_SM_P version 3 data products. SMAP Project,JPL D-93720. Jet Propulsion Laboratory, Pasadena, CA.

Kerr, Y., 2007. Soil moisture from space: where are we? Hydrogeol. J. 15, 117–120.Kerr, Y., Jacquette, E., Al Bitar, A., Cabot, F., Mialon, A., Richaume, P., Quesney, A., Berthon,

L., Wigneron, J., 2013. The CATDS SMOS L3 Soil Moisture Retrieval Processor, Algo-rithm Theoretical Baseline Document (ATBD). SO-TN-CBSA-GS-0029.

Kerr, Y.H., Waldteufel, P., Wigneron, J.P., Martinuzzi, J., Font, J., Berger, M., 2001. Soil mois-ture retrieval from space: the soil moisture and ocean salinity (SMOS) mission. IEEETrans. Geosci. Remote Sens. 39, 1729–1735.

Kerr, Y.H., Waldteufel, P., Wigneron, J.P., Delwart, S., Cabot, F., Boutin, J., Escorihuela, M.J.,Font, J., Reul, N., Gruhier, C., Juglea, S.E., Drinkwater, M.R., Hahne, A., Martin-Neira, M.,Mecklenburg, S., 2010. The SMOS mission: new tool for monitoring key elementsofthe global water cycle. Proc. IEEE 98, 666–687.

Kerr, Y.H., Waldteufel, P., Richaume, P., Wigneron, J.P., Ferrazzoli, P., Mahmoodi, A., AlBitar, A., Cabot, F., Gruhier, C., Juglea, S.E., Leroux, D., Mialon, A., Delwart, S., 2012.The SMOS soil moisture retrieval algorithm. IEEE Trans. Geosci. Remote Sens. 50,1384–1403.

Kerr, Y.H., Al-Yaari, A., Rodriguez-Fernandez, N., Parrens, M., Molero, B., Leroux, D.,Bircher, S., Mahmoodi, A., Mialon, A., Richaume, P., Delwart, S., Al Bitar, A., Pellarin,T., Bindlish, R., Jackson, T.J., Rüdiger, C., Waldteufel, P., Mecklenburg, S., Wigneron,J.P., 2016. Overview of SMOS performance in terms of global soil moisturemonitoringafter six years in operation. Remote Sens. Environ. 180, 40–63.

Kolassa, J., Aires, F., Polcher, J., Prigent, C., Jimenez, C., Pereira, J.M., 2013. Soil moisture re-trieval from multi-instrument observations: information content analysis and re-trieval methodology. J. Geophys. Res. Atmos. 118, 4847–4859.

Larson, K.M., Small, E.E., Gutmann, E.D., Bilich, A.L., Braun, J.J., Zavorotny, V.U., 2008. Use ofGPS receivers as a soil moisture network for water cycle studies. Geophys. Res. Lett.35 (n/a-n/a).

Liu, Y.Y., Dorigo, W.A., Parinussa, R.M., de Jeu, R.A.M., Wagner, W., McCabe, M.F., Evans,J.P., van Dijk, A.I.J.M., 2012. Trend-preserving blending of passive and active micro-wave soil moisture retrievals. Remote Sens. Environ. 123, 280–297.

Lv, S., Zeng, Y., Wen, J., Su, Z., 2016. A reappraisal of global soil effective temperatureschemes. Remote Sens. Environ. 183, 144–153.

Mattar, C., Wigneron, J.P., Sobrino, J.A., Novello, N., Calvet, J.C., Albergel, C., Richaume, P.,Mialon, A., Guyon, D., Jimenez-Munoz, J.C., Kerr, Y., 2012. A combined optical & mi-crowave method to retrieve soil moisture over vegetated areas. IEEE Trans. Geosci.Remote Sens. 50, 1404–1413.

Mo, T., Choudhury, B.J., Schmugge, T.J., Wang, J.R., Jackson, T.J., 1982. A model for micro-wave emission from vegetation-covered fields. J. Geophys. Res. Oceans 87,11229–11237.

Njoku, E.G., Li, L., 1999. Retrieval of land surface parameters using passive microwavemeasurements at 6–18 GHz. IEEE Trans. Geosci. Remote Sens. 37, 79–93.

Njoku, E.G.,Wilson,W.J., Yueh, S.H., Dinardo, S.J., Fuk, K.L., Jackson, T.J., Lakshmi, V., Bolten,J., 2002. Observations of soil moisture using a passive and active low-frequency mi-crowave airborne sensor during SGP99. IEEE Trans. Geosci. Remote Sens. 40,2659–2673.

Njoku, E.G., Jackson, T.J., Lakshmi, V., Chan, T.K., Nghiem, S.V., 2003. Soil moisture retrievalfrom AMSR-E. IEEE Trans. Geosci. Remote Sens. 41, 215–229.

Oliva, R., Daganzo, E., Kerr, Y.H., Mecklenburg, S., Nieto, S., Richaume, P., Gruhier, C., 2012.SMOS radio frequency interference scenario: status and actions taken to improve theRFI environment in the 1400–1427-MHz passive band. IEEE Trans. Geosci. RemoteSens. 50, 1427–1439.

Oliva, R., Daganzo, E., Richaume, P., Kerr, Y., Cabot, F., Soldo, Y., Anterrieu, E., Reul, N.,Gutierrez, A., Barbosa, J., Lopes, G., 2016. Status of radio frequency interference(RFI) in the 1400–1427 MHz passive band based on six years of SMOS mission. Re-mote Sens. Environ. 180, 64–75.

Owe, M., Van de Griend, A.A., 1998. Comparison of soil moisture penetration depths forseveral bare soils at two microwave frequencies and implications for remote sensing.Water Resour. Res. 34, 2319–2327.

Owe, M., de Jeu, R., Walker, J., 2001. A methodology for surface soil moisture and vegeta-tion optical depth retrieval using the microwave polarization difference index. IEEETrans. Geosci. Remote Sens. 39, 1643–1654.

Parinussa, R.M., Holmes, T.R.H., Yilmaz, M.T., Crow, W.T., 2011. The impact of land surfacetemperature on soil moisture anomaly detection from passive microwave observa-tions. Hydrol. Earth Syst. Sci. 15, 3135–3151.

Parrens, M., Zakharova, E., Lafont, S., Calvet, J.C., Kerr, Y., Wagner, W., Wigneron, J.P., 2012.Comparing soil moisture retrievals from SMOS and ASCAT over France. Hydrol. EarthSyst. Sci. 16, 423–440.

Parrens, M., Wigneron, J.-P., Richaume, P., Mialon, A., Al Bitar, A., Fernandez-Moran, R., Al-Yaari, A., Kerr, Y.H., 2016. Global-scale surface roughness effects at L-band as estimat-ed from SMOS observations. Remote Sens. Environ. 181, 122–136.

Piepmeier, J., Mohammed, P., De Amici, G., Kim, E., Peng, J., Ruf, C., Hanna, M., Yueh, S.,Entekhabi, D., 2016. Soil Moisture Active Passive (SMAP) Project Algorithm Theoret-ical Basis Document SMAP L1B Radiometer Data Product: L1B_TB.

Rautiainen, K., Lemmetyinen, J., Pulliainen, J., Vehvilainen, J., Drusch, M., Kontu, A.,Kainulainen, J., Seppanen, J., 2012. L-band radiometer observations of soil processesin boreal and subarctic environments. IEEE Trans. Geosci. Remote Sens. 50, 1483–1497.

Reichle, R., Draper, C., Koster, R., Liu, A., Girotto, M., Mahanama, S., De Lannoy, G., 2016.Assessment of MERRA-2 land surface hydrology estimates. J. Climatol. (submitted).

Reichle, R.H., Koster, R.D., De Lannoy, G.J.M., Forman, B.A., Liu, Q., Mahanama, S.P.P., Touré,A., 2011. Assessment and enhancement of MERRA land surface hydrology estimates.J. Clim. 24, 6322–6338.

Rodríguez-Fernández, J.N., Kerr, H.Y., van der Schalie, R., Al-Yaari, A., Wigneron, J.-P., deJeu, R., Richaume, P., Dutra, E., Mialon, A., Drusch, M., 2016. Long term global surfacesoil moisture fields using an SMOS-trained neural network applied to AMSR-E data.Remote Sens. 8.

Rondinelli, W.J., Hornbuckle, B.K., Patton, J.C., Cosh, M.H., Walker, V.A., Carr, B.D., Logsdon,S.D., 2015. Different rates of soil drying after rainfall are observed by the SMOS satel-lite and the south fork in situ soil moisture network. J. Hydrometeorol. 16, 889–903.

Saleh, K., Wigneron, J.-P., de Rosnay, P., Calvet, J.-C., Kerr, Y., 2006. Semi-empirical regres-sions at L-band applied to surface soil moisture retrievals over grass. Remote Sens.Environ. 101, 415–426.

Sanchez, N., Martinez-Fernandez, J., Scaini, A., Perez-Gutierrez, C., 2012. Validation of theSMOS L2 soil moisture data in the REMEDHUS network (Spain). IEEE Trans. Geosci.Remote Sens. 50, 1602–1611.

273A. Al-Yaari et al. / Remote Sensing of Environment 193 (2017) 257–273

Santamaría-Artigas, A., Mattar, C., Wigneron, J.P., 2016. Application of a combined optical–passive microwave method to retrieve soil moisture at regional scale over Chile. IEEEJ. Sel. Top. Appl. Earth Obs. Remote Sens. 9, 1493–1504.

Schaefer, G.L., Cosh, M.H., Jackson, T.J., 2007. The USDA natural resources conservation ser-vice soil climate analysis network (SCAN). J. Atmos. Ocean. Technol. 24, 2073–2077.

Schmugge, T., Wilheit, T., Webster Jr., W., Gloersen, P., 1976. Remote sensing of soil mois-ture with microwave radiometers-II. Nat. Aeronautics and Space Admin., NASA Tech.Note D-8321.

Seneviratne, S.I., Corti, T., Davin, E.L., Hirschi, M., Jaeger, E.B., Lehner, I., Orlowsky, B.,Teuling, A.J., 2010. Investigating soil moisture–climate interactions in a changing cli-mate: a review. Earth Sci. Rev. 99, 125–161.

Smith, A.B., Walker, J.P., Western, A.W., Young, R.I., Ellett, K.M., Pipunic, R.C., Grayson, R.B.,Siriwardena, L., Chiew, F.H.S., Richter, H., 2012. The Murrumbidgee soil moisturemonitoring network data set. Water Resour. Res. 48, W07701.

Tagesson, T., Fensholt, R., Guiro, I., Rasmussen, M.O., Huber, S., Mbow, C., Garcia, M.,Horion, S., Sandholt, I., Holm-Rasmussen, B., Göttsche, F.M., Ridler, M.-E., Olén, N.,Lundegard Olsen, J., Ehammer, A., Madsen, M., Olesen, F.S., Ardö, J., 2015. Ecosystemproperties of semiarid savanna grassland inWest Africa and its relationship with en-vironmental variability. Glob. Chang. Biol. 21, 250–264.

Taylor, K.E., 2001. Summarizing multiple aspects of model performance in a single dia-gram. J. Geophys. Res. Atmos. 106, 7183–7192.

Ulaby, F.T., Dubois, P.C., van Zyl, J., 1996. Radar mapping of surface soil moisture. J. Hydrol.184, 57–84.

Van der Schalie, R., Kerr, Y.H., Wigneron, J.P., Rodríguez-Fernández, N.J., Al-Yaari, A., Jeu,R.A.M.D., 2016. Global SMOS soil moisture retrievals from the land parameter retriev-al model. Int. J. Appl. Earth Obs. Geoinf. 45 (Part B), 125–134.

Vittucci, C., Ferrazzoli, P., Kerr, Y., Richaume, P., Guerriero, L., Rahmoune, R., Laurin, G.V.,2016. SMOS retrieval over forests: exploitation of optical depth and tests of soil mois-ture estimates. Remote Sens. Environ. 180, 115–127.

Wagner, W., Dorigo, W., de Jeu, Richard, Fernandez, Diego, Benveniste, Jerome, Haas, Eva,Ertl, M., 2012. Fusion of Active and Passive Microwave Observations to Create an Es-sential Climate Variable Data Record on SoilMoisture. XII Congress of the Internation-al Society for Photogrammetry and Remote Sensing-Melbourne, Australia.

Wang, J.R., Choudhury, B.J., 1981. Remote sensing of soil moisture content, over bare fieldat 1.4 GHz frequency. J. Geophys. Res. Oceans 86, 5277–5282.

Wang, J.R., O'Neill, P.E., Jackson, T.J., Engman, E.T., 1983. Multifrequency measurements ofthe effects of soil moisture, soil texture, and surface roughness. IEEE Trans. Geosci. Re-mote Sens. GE-21, 44–51.

Western, A.W., Zhou, S.-L., Grayson, R.B., McMahon, T.A., Blöschl, G., Wilson, D.J., 2004.Spatial correlation of soil moisture in small catchments and its relationship to domi-nant spatial hydrological processes. J. Hydrol. 286, 113–134.

Wigneron, J.-P., Chanzy, A., Calvet, J.-C., Bruguier, N., 1995. A simple algorithm to retrievesoil moisture and vegetation biomass using passive microwave measurements overcrop fields. Remote Sens. Environ. 51, 331–341.

Wigneron, J.P., Calvet, J.C., Pellarin, T., Van de Griend, A.A., Berger, M., Ferrazzoli, P., 2003.Retrieving near-surface soil moisture frommicrowave radiometric observations: cur-rent status and future plans. Remote Sens. Environ. 85, 489–506.

Wigneron, J.P., Calvet, J.C., De Rosnay, P., Kerr, Y., Waldteufel, P., Saleh, K., Escorihuela, M.J.,Kruszewski, A., 2004. Soil moisture retrievals from biangular L-band passive micro-wave observations. IEEE Geosci. Remote Sens. Lett. 1, 277–281.

Wigneron, J.P., Kerr, Y., Waldteufel, P., Saleh, K., Escorihuela, M.J., Richaume, P., Ferrazzoli,P., de Rosnay, P., Gurney, R., Calvet, J.C., Grant, J.P., Guglielmetti, M., Hornbuckle, B.,Mätzler, C., Pellarin, T., Schwank, M., 2007. L-band microwave emission of the bio-sphere (L-MEB) model: description and calibration against experimental data setsover crop fields. Remote Sens. Environ. 107, 639–655.

Wigneron, J.P., Chanzy, A., Kerr, Y.H., Lawrence, H., Shi, J., Escorihuela, M.J., Mironov, V.,Mialon, A., Demontoux, F., Rosnay, P.D., Saleh-Contell, K., 2011. Evaluating an im-proved parameterization of the soil emission in L-MEB. IEEE Trans. Geosci. RemoteSens. 49, 1177–1189.

Wigneron, J.P., Jackson, T.J., O'Neill, P., De Lannoy, G., de Rosnay, P., Walker, J.P., Ferrazzoli,P., Mironov, V., Bircher, S., Grant, J.P., Kurum, M., Schwank, M., Munoz-Sabater, J., Das,N., Royer, A., Al-Yaari, A., Al Bitar, A., Fernandez-Moran, R., Lawrence, H., Mialon, A.,Parrens, M., Richaume, P., Delwart, S., Kerr, Y., 2017. Modelling the passive micro-wave signature from land surfaces: A review of recent results and application tothe L-band SMOS & SMAP soil moisture retrieval algorithms. Remote Sens. Environ.192, 238–262.

Zeng, J., Chen, K.S., Bi, H., Chen, Q., 2016. A preliminary evaluation of the SMAP radiometersoil moisture product over United States and Europe using ground-based measure-ments. IEEE Trans. Geosci. Remote Sens. (PP, 1-0).